Monte-Carlo Pricing under a Hybrid Local Volatility...

Monte-Carlo Pricing under a Hybrid Local

Volatility model

Sebastien Gurrieri

Mizuho International plc

GPU Technology ConferenceSan Jose, 14-17 May 2012

Sebastien Gurrieri Hybrid Local Volatility in Monte-Carlo

Learn how to create a scalable volume visualization system for interactive rendering of terascale EM data. Presented at the GPU Technology Conference 2012.

Introduction

Key Interests in Finance

Pricing of exotic derivatives

Monte-Carlo simulations

Local Volatility model for Foreign Exchange Rates (FX)

Hybrid with Interest Rate models (IR)


Introduction

Key Interests in Finance

Pricing of exotic derivatives

Monte-Carlo simulations

Local Volatility model for Foreign Exchange Rates (FX)

Hybrid with Interest Rate models (IR)

Key Interests in CUDA

High-dimensional Monte-Carlo simulations

Texture memory (layered)


Introduction

Plan of the talk

Description of the problem and motivation for parallelprogramming and textures


Introduction

Plan of the talk


Outline of implementation in CUDA


Introduction

Plan of the talk



Numerical tests

Call/Put options in Local Volatility (LV) modelExotic swaps in LV modelExotic swaps in Hybrid LV model


Introduction

Plan of the talk



Numerical tests

Call/Put options in Local Volatility (LV) modelExotic swaps in LV modelExotic swaps in Hybrid LV model

Conclusion on performance and use in industry


Description of the problem

The product: Power-Reverse Dual Coupon Swap (PRDC)




Underlying swap: for a series of dates 0 ≤ Ti ≤ 30 years





receive option on FXi with strike Ki :

+max(FXi − Ki , 0)





receive option on FXi with strike Ki :

+max(FXi − Ki , 0)

pay option on IRi with strike Qi :

−max(IRi − Qi , 0)



The product: Exotic exercise

Target Redemption Note (TARN) with target A





Monitor coupon sum

Ci =

i∑

k=1

max(FXk − Kk , 0)





Monitor coupon sum

Ci =

i∑

k=1

max(FXk − Kk , 0)

if Ci > A, cancel all remaining cash-flows



The product: Main features

Sensitive to FX smile

→ modelling of smile






Sensitive to FX-IR correlation, IR volatility

→ modelling of IR stochasticity

→ multi-factor FX-IR hybrid






Sensitive to FX-IR correlation, IR volatility

→ modelling of IR stochasticity

→ multi-factor FX-IR hybrid

Path-dependent due to exotic exercise

→ mainly Monte-Carlo



The model: Dupire’s Local Volatility [1]

Diffusion with volatility σ(t,FX )

dFX

FX= (rd − rf )dt + σ(t,FX )dW

rd is the domestic interest rate

rf is the foreign interest rate

dW is a Brownian motion



The model: Calibration to the market of FX options




Market characterized by implied volatility θ(t,FX )

→ once differentiable in t, twice in FX (ideally)

→ satisfies non-arbitrage conditions (ideally)




Market characterized by implied volatility θ(t,FX )

→ once differentiable in t, twice in FX (ideally)

→ satisfies non-arbitrage conditions (ideally)

Model fits the market exactly for Dupire’s condition

σ2(t,FX ) = f(∂θ

∂t,∂θ

∂FX,∂2θ

∂FX 2

)



The model: Sampling the volatility

LV matrix defined as

σni = σ(tn,FXi)





σni = σ(tn,FXi)

Typical size ∼ 200× 200 = 40, 000 entries





σni = σ(tn,FXi)


Bi-linear interpolation in t and FX

→ texture memory [2]

→ simple but lacks flexibility





σni = σ(tn,FXi)


Bi-linear interpolation in t and FX

→ texture memory [2]

→ simple but lacks flexibility

Linear interpolation in FX at known t

→ layered textures

→ slightly more complicated but more flexible and/oraccurate



Summary

Multi-factor and path-dependent product

→ Monte-Carlo simulation

→ good speed-up expected with CUDA



Summary




Model requires interpolation of a matrix

→ benefit from texture memory



Summary




Model requires interpolation of a matrix

→ benefit from texture memory

Multiple cash-flows, monitoring, smile-modelling

→ large number of time steps

→ high-dimensional problem

→ inline random number generation


Implementation Outline

Single-thread:

• On each path j , at each time tn



Single-thread:


1 calculate next uniform random number



Single-thread:



2 transform to Gaussian, then Brownian motion increment dW jn



Single-thread:




3 read previous spot FX jn from memory



Single-thread:





4 calculate volatility σ by calling texture at (tn,FXjn)



Single-thread:






5 calculate new spot

FXjn+1 = FX j

ne(rd−rf −

12σ2)(tn+1−tn)+σdW

jn



Single-thread:







FXjn+1 = FX j

ne(rd−rf −


jn

6 calculate product(s)



Single-thread:







FXjn+1 = FX j

ne(rd−rf −


jn


7 write new spot in memory



Single-thread:







FXjn+1 = FX j

ne(rd−rf −


jn


7 write new spot in memory

• Loop on path, then time.



Multi-thread:

Sequential in time, parallel on paths



Multi-thread:


Grid configuration

1-dimensional grid of Nblocks blocks1-dimensional blocks of Nthreads threadss = Nblocks × Nthreads = number of concurrent threads



Multi-thread:


Grid configuration


Thread j calculates paths j , j + s, j + 2s, etc ...



Multi-thread:


Grid configuration


Thread j calculates paths j , j + s, j + 2s, etc ...

Thread j must remember previous spot values for pathsj , j + s, j + 2s, etc ...

→ too much for shared memory

→ store previous spot values in global memory



Multi-thread:

Thread j :

calculates products on paths j , j + s, j + 2s, etc ...sums them in local variablewrites sums in shared memory



Multi-thread:

Thread j :


Synchronize



Multi-thread:

Thread j :


Synchronize

In each block:

one thread is attributed to each productaccumulates in a local variable all thread sums for this productwrites ”block-partial” sum in global memory



Multi-thread:

Global memory contains ”block-partial” sums for eachproduct, each block, at each time



Multi-thread:


Transfer to host



Multi-thread:


Transfer to host

On host, sum results of all blocks.



Multi-thread: remark on random number generation

typical number of times: 500





typical number of factors: 2, but easily going to 3 and more






typical number of simulations: 100K, but may want more







global generation requires minimum global memory

500× 2× 100, 000 × 4 = 400MB







global generation requires minimum global memory

500× 2× 100, 000 × 4 = 400MB

cannot run on all devices, too restrictive for praticalapplications

→ use inline generation



Texture:

Desired interpolation



Texture:

Texture interpolation



Texture:

Linear rescaling is required



Texture:


Given spots FX0,FX1, · · · FXM−1, volatilities σ0, σ1, · · · σM−1



Texture:



The volatility at any spot FX is

σ(FX ) = tex(αFX + β)



Texture:



The volatility at any spot FX is

σ(FX ) = tex(αFX + β)

with

α =M − 1

M(FXM−1 − FX0)

β =1

M

(1

2− (M − 1)

FX0

FXM−1 − FX0

)



Texture:

Bi-linear interpolation with standard texture

σ(t,FX ) = tex2D(αFX + β, γt + δ)



Texture:

Bi-linear interpolation with standard texture

σ(t,FX ) = tex2D(αFX + β, γt + δ)

Linear interpolation with layered texture

σ(tn,FX ) = tex1DLayered(αFX + β, n)


Numerical Tests

Vanilla Options:

Performance of the texture (500 time steps, 500K simulations)

50% ∼ 70% speed gains with texture

good accuracy of the texture interpolation

∼ 100 points sufficient


Numerical Tests

Vanilla Options:

Gain (single thread vs. GTX 460)


Numerical Tests

Exotic Swap (one factor):

Additional state variable on path j

Cji =

i∑

k=1

max(FX jk − Kk , 0)

→ one more read/write access from global memory


Numerical Tests


Additional state variable on path j

Cji =

i∑

k=1

max(FX jk − Kk , 0)

→ one more read/write access from global memory

Product calculated only at cash-flow times (at most 120)

→ less operations than for vanillas (500)


Numerical Tests




Numerical Tests

Exotic Swap (hybrid 2F):

rd follows Hull-White model

drd = (θ − ard )dt + σrdWr


Numerical Tests




it has a correlation ρ with FX

dWFX = g1√dt

dWr = (ρg1 +√

1− ρ2g2)√dt


Numerical Tests




it has a correlation ρ with FX

dWFX = g1√dt

dWr = (ρg1 +√

1− ρ2g2)√dt

2 additional state variable on path j

r j , e∫ T

0r jdt (numeraire)

→ two more read/write accesses to global memory


Numerical Tests




Conclusion

Extention to 3F, barriers

→ very similar to 2F, TARN

→ should not be a problem


Conclusion

Extention to 3F, barriers

→ very similar to 2F, TARN

→ should not be a problem

Extention to callables

→ Longstaff-Schwartz not entirely parallel

→ should not be a problem but gains may be lower

→ use Malliavin calculus? (Abbas-Turki, GTC 2010)


Conclusion

Large gains on GTX 460 and for realistic products and pricingconfigurations


Conclusion


Possibility to run more simulations

→ more accurate Greeks

→ more efficient risk management


Conclusion





Value-at-Risk and Potential Exposure calculations possiblewithout approximations


Conclusion





Value-at-Risk and Potential Exposure calculations possiblewithout approximations

Large number of scenario testing possible on exotic portfolios


Disclaimer

This publication has been prepared by Sebastien Gurrieri of Mizuho International plcsolely for the purpose of presentation at this conference. The opinions expressed inthis presentation are those of the author and do not necessarily reflect the view ofMizuho International plc, which is not responsible for any use which may be made ofits contents.

It is not, and should not be construed as, an offer or solicitation to buy, or sell, anysecurity, or any interest in a security or enter into any transaction. This publicationmay include details of instruments that have not been issued. There is no guaranteethat such instruments will be issued in the future.

This publication has been prepared solely from publicly available information.Information contained herein and the data underlying it have been obtained from, orbased upon, sources believed by the author to be reliable. However, no assurance canbe given that the information, data or any computations based thereon, is accurate orcomplete. Opinions stated in this report are subject to change without notice.There are risks associated with the financial instruments and transactions described inthis publication. Investors should consult their own financial, legal, accounting and taxadvisors about the risks, the appropriate tools to analyse an investment and thesuitability of an investment in their particular circumstances. Mizuho International plcis not responsible for assessing the suitability of any investment. Investment decisionsand responsibility for any investments is the sole responsibility of the investor. Neitherthe author, Mizuho International plc nor any affiliate accepts any liability whatsoeverwith respect to the use of this report or its contents.


References

1 B. Dupire, ”Pricing with a smile,” Risk 7, pp. 18-20, Jan.1994.

2 A. Bernemann, R. Schreyer and K. Spanderen, ”AcceleratingExotic Option Pricing and Model Calibration Using GPUs”,Working Paper, Feb. 2011.

3 http://sebgur.fr/sgdev.html


Monte-Carlo Pricing under a Hybrid Local Volatility...

Documents

Transcript of Monte-Carlo Pricing under a Hybrid Local Volatility...